Abstract:

In the realm of cotton disease identification, the Deep Neural Network emerges as a pivotal paradigm. Progress in this sphere hinges on the availability of a comprehensive repository of scientific data, encapsulating a broader spectrum of diseases, variegated soil profiles, and multifaceted environmental attributes. Currently, this dearth of data serves as the principal bottleneck, impeding the advancement of state-of-the-art models and algorithms.Within this scholarly exposition, we present a meticulously curated cotton disease dataset, poised to bridge this knowledge chasm. This dataset comprehensively encompasses four prevalent cotton diseases: anthracnose, bacterial blight, brown spot, and wilt disease. These maladies' exemplars were meticulously gleaned from cotton fields situated in the Potianyang High-standard Farmland Demonstration Base, nestled serenely in Sanya, Hainan Province, China.The dataset unfolds as a magnum opus, comprising 3 453 high-resolution images. These vivid snapshots provide a panoramic view, capturing the pristine vitality of healthy leaves, juxtaposed with leaves beset by disease at various growth stages. The data acquisition, executed with precision, leveraged field random sampling methodologies, ensuring a faithful reflection of the natural complexity in real-world cotton plantations.Every image underwent meticulous scrutiny, with ten seasoned mavens in cotton pathology meticulously overseeing the annotation. An additional cohort of twenty annotators conducted a second round of annotations on randomly selected image subsets, fortifying the dataset's integrity and precision. The Vision Transformer model was employed to guarantee the dataset's resilience and accuracy.This hallowed dataset was meticulously gathered amidst the complexity of field environments, encapsulating the nuances of major cotton diseases in their native habitat. Its high image resolution, akin to an opulent tapestry of visual data, renders it an invaluable resource for pioneering research, astute training, and the relentless validation of astute, intelligent cotton disease recognition models and algorithms. This opulent repository caters to the discriminating tastes of researchers, practitioners, and sagacious decision-makers, furnishing them with a comprehensive and crystalline understanding of the multifaceted tapestry of cotton diseases and their intricate management.

Key words: deep learning, cotton anthracnose, cotton angular leaf spot, cotton brown spot, Fusarium wilt, image recognition technology